A method executed by a processor may include storing account activity data in a database which identifies prior account activities performed by customers in their respective accounts associated with a financial institution. The database may be accessible only by computing systems of the financial institution. A set of challenge questions may be received for authenticating customers to perform high-risk activities in their respective accounts that are based on prior account activities. A machine learning model may be used to determine an authentication score used for ranking each challenge question in the set of challenge questions. An electronic request may be received from an unverified customer who desires to perform high-risk activities in an account. Challenge questions may be selected based on the ranking. The unverified customer may be verified when correctly answering the selected challenge questions. The verified customer may be allowed to perform high-risk activities in the account.
Legal claims defining the scope of protection, as filed with the USPTO.
1. A method comprising: storing, by a processor, in a database in a computer memory, account activity data identifying prior account activities performed by a plurality of customers in their respective accounts associated with a financial institution; wherein the database stored in the computer memory is accessible only by computing systems of the financial institution; receiving, by the processor, a set of challenge questions for authenticating the plurality of customers to perform high-risk activities in their respective accounts; wherein the set of challenge questions are based on the prior account activities performed by the plurality of customers in their respective accounts; utilizing, by the processor, at least one machine learning model to determine an authentication score for each challenge question in the set of challenge questions; wherein the authentication score of each respective challenge question is based on: i) a first number of instances of a plurality of customer interactions with the financial institution that a correct answer to a respective challenge question in the set of challenge questions has been given by the plurality of customers of the financial institution, ii) a second number of instances of the plurality of customer interactions with the financial institution that the correct answer to the respective challenge question in the set of challenge questions has been given by at least one fraudulent person, iii) a third number of instances of the plurality of customer interactions with the financial institution that an incorrect answer to the respective challenge question in the set of challenge questions has been given by the plurality of customers of the financial institution, and iv) a fourth number of instances of the plurality of customer interactions with the financial institution that the incorrect answer to the respective challenge question in the set of challenge questions has been given by the at least one fraudulent person; identifying, by the processor, the at least one fraudulent person from the plurality of customers by tagging customer interactions in the plurality of customer transactions as fraudulent; training, by the processor, the at least one machine learning model with training data for each challenge question in the set of challenge questions for each respective customer interaction from the plurality of customer interactions; wherein the training data comprises: (iii) a first indication that a correct or an incorrect answer was given for each challenge question in the set of challenge questions for each respective customer interaction from the plurality of customer interactions, and (iv) a second indication of a fraud tag applied to each respective customer interaction from the plurality of customer interactions; ranking, by the processor, challenge questions of the set of challenge questions from a highest authentication score to a lowest authentication score; receiving, by the processor, an electronic request on a computing device from an unverified customer who desires to perform one or more high-risk activities in an account of a particular customer of the plurality of customers; selecting, by the processor, a predefined number of challenge questions having the highest authentication scores based on the ranking; causing, by the processor, to display on a screen of the computing device the selected challenge questions; receiving, by the processor, answers to the selected challenge questions; authenticating, by the processor, the unverified customer to form a verified customer when the answers to the selected challenge questions are correct; and allowing, by the processor, the verified customer from the plurality of customers to perform the one or more high-risk activities with a respective account associated with the verified customer.
2. The method according to claim 1 , wherein identifying the at least one fraudulent person from the plurality of customers by tagging customer interactions in the plurality of customer transactions as fraudulent comprises receiving a customer self-report that a customer interaction from the plurality of customer transactions with the financial institution is fraudulent.
3. The method according to claim 1 , wherein identifying the at least one fraudulent person from the plurality of customers by tagging customer interactions in the plurality of customer transactions as fraudulent comprises receiving a notification from a fraud department of the financial institution indicating that a customer interaction from the plurality of customer transactions with the financial institution is fraudulent.
4. The method according to claim 1 , wherein identifying the at least one fraudulent person from the plurality of customers by tagging customer interactions in the plurality of customer transactions as fraudulent comprises assessing that the unverified customer communicated with the financial institution through an IP address or a telephone number previously associated with fraudulent activity.
5. The method according to claim 1 , wherein the high-risk activities in the respective accounts of the plurality of customers are selected from the group consisting of: a change of an account address, a change of an account e-mail address, a change in a cellphone number associated with an account, a change in a telephone number associated with an account, a balance transfer request, and a request to have a credit card sent to a new address.
6. The method according to claim 1 , wherein receiving answers to the selected challenge questions is selected from the group consisting of: entering the answers of the unverified customer by an agent of the financial institution into a computer terminal of the agent while communicating with the unverified customer, receiving the answers through a webpage of the financial institution from the unverified user, and receiving the answers through a mobile application of the financial institution from the unverified user.
7. The method according to claim 1 , wherein the at least one machine learning model is selected from the group consisting of a multi-armed bandit model and a multiclass classifier neural network model.
8. The method according to claim 1 , wherein training the at least one machine learning model with the training data comprises training the machine learning model at predefined time intervals with new training data.
9. The method according to claim 1 , wherein training the at least one machine learning model with the training data comprises applying a loss function to the training data.
10. The method according to claim 1 , wherein training the at least one machine learning model with the training data comprises splitting the training data for each respective customer interaction from the plurality of customer interactions into two sets of training data with a first set for digitally savvy customers and a second set for non-digitally savvy customers.
11. A system, comprising: a computer memory; and a processor configured to: store in a database in the computer memory, account activity data identifying prior account activities performed by a plurality of customers in their respective accounts associated with a financial institution; wherein the database stored in the computer memory is accessible only by computing systems of the financial institution; receive a set of challenge questions for authenticating the plurality of customers to perform high-risk activities in their respective accounts; wherein the set of challenge questions are based on the prior account activities performed by the plurality of customers in their respective accounts; utilize at least one machine learning model to determine an authentication score for each challenge question in the set of challenge questions; wherein the authentication score of each respective challenge question is based on: i) a first number of instances of a plurality of customer interactions with the financial institution that a correct answer to a respective challenge question in the set of challenge questions has been given by the plurality of customers of the financial institution, ii) a second number of instances of the plurality of customer interactions with the financial institution that the correct answer to the respective challenge question in the set of challenge questions has been given by at least one fraudulent person, iii) a third number of instances of the plurality of customer interactions with the financial institution that an incorrect answer to the respective challenge question in the set of challenge questions has been given by the plurality of customers of the financial institution, and iv) a fourth number of instances of the plurality of customer interactions with the financial institution that the incorrect answer to the respective challenge question in the set of challenge questions has been given by the at least one fraudulent person; identify the at least one fraudulent person from the plurality of customers by tagging customer interactions in the plurality of customer transactions as fraudulent; train the at least one machine learning model with training data for each challenge question in the set of challenge questions for each respective customer interaction from the plurality of customer interactions; wherein the training data comprises: (i) a first indication that a correct or an incorrect answer was given for each challenge question in the set of challenge questions for each respective customer interaction from the plurality of customer interactions, and (ii) a second indication of a fraud tag applied to each respective customer interaction from the plurality of customer interactions; rank challenge questions of the set of challenge questions from a highest authentication score to a lowest authentication score; receive an electronic request on a computing device from an unverified customer who desires to perform one or more high-risk activities in an account of a particular customer of the plurality of customers; select a predefined number of challenge questions having the highest authentication scores based on the ranking; cause to display on a screen of the computing device the selected challenge questions; receive answers to the selected challenge questions; authenticate the unverified customer to form a verified customer when the answers to the selected challenge questions are correct; and allow the verified customer from the plurality of customers to perform the one or more high-risk activities with a respective account associated with the verified customer.
12. The system according to claim 11 , wherein the processor is configured to identify the at least one fraudulent person from the plurality of customers by tagging customer interactions in the plurality of customer transactions as fraudulent by receiving a customer self-report that a customer interaction from the plurality of customer transactions with the financial institution is fraudulent.
13. The system according to claim 11 , wherein the processor is configured to identify the at least one fraudulent person from the plurality of customers by tagging customer interactions in the plurality of customer transactions as fraudulent by receiving a notification from a fraud department of the financial institution indicating that a customer interaction from the plurality of customer transactions with the financial institution is fraudulent.
14. The system according to claim 11 , wherein the processor is configured to identify the at least one fraudulent person from the plurality of customers by tagging customer interactions in the plurality of customer transactions as fraudulent by assessing that the unverified customer communicated with the financial institution through an IP address or a telephone number previously associated with fraudulent activity.
15. The system according to claim 11 , wherein the high-risk activities in the respective accounts of the plurality of customers are selected from the group consisting of: a change of an account address, a change of an account e-mail address, a change in a cellphone number associated with an account, a change in a telephone number associated with an account, a balance transfer request, and a request to have a credit card sent to a new address.
16. The system according to claim 11 , wherein the processor is configured to receive answers to the selected challenge questions is selected from the group consisting of entering the answers of the unverified customer by an agent of the financial institution into a computer terminal of the agent while communicating with the unverified customer, receiving the answers through a webpage of the financial institution from the unverified user, and receiving the answers through a mobile application of the financial institution from the unverified user.
17. The system according to claim 11 , wherein the at least one machine learning model is selected from the group consisting of a multi-armed bandit model and a multiclass classifier neural network model.
18. The system according to claim 11 , wherein the processor is configured to train the at least one machine learning model with the training data by training the machine learning model at predefined time intervals with new training data.
19. The system according to claim 11 , wherein the processor is configured to train the at least one machine learning model with the training data by applying a loss function to the training data.
20. The system according to claim 11 , wherein the processor is configured to train the at least one machine learning model with the training data by splitting the training data for each respective customer interaction from the plurality of customer interactions into two sets of training data with a first set for digitally savvy customers and a second set for non-digitally savvy customers.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
September 9, 2019
February 25, 2020
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.